A Bayesian and efficient observer model explains concurrent attractive and repulsive history biases in visual...

Area of Science:

• Cognitive Neuroscience
• Computational Vision
• Psychophysics

Background:

• Human perception exhibits history-dependent biases, including attractive serial dependence and repulsive adaptation.
• The interaction and underlying computational principles of these opposing biases remain unclear.

Purpose of the Study:

• To disentangle attractive and repulsive perceptual biases by examining their distinct timescales.
• To elucidate the computational mechanisms governing these history dependencies in visual processing.

Main Methods:

• Investigated the temporal dynamics of attractive and repulsive biases in human perceptual decisions.
• Utilized computational modeling, including ideal observer models with efficient encoding and Bayesian decoding.

Main Results:

• Perceptual decisions demonstrated concurrent attraction to short-term history and repulsion from stimuli experienced minutes earlier.
• This temporal pattern was not explained by a standard ideal Bayesian observer model alone.
• An ideal observer model incorporating efficient encoding and Bayesian decoding in a slowly changing environment accurately captured the observed biases.

Conclusions:

• Concurrent attractive and repulsive history biases reflect a dual constraint in visual processing.
• These biases likely arise from the need to balance perceptual efficiency with environmental stability.